LLM Digest
Subscribe

AI Daily Recap

25 articles · 6 categories

View as JSON
‹›

The finishable daily brief

What happened in AI — Nov 16, 2026

Monday, Nov 16, 2026
25 articles · 6 categories

read top to bottom · then stop

In 30 seconds

  • Claude Code team: ~90% of its code is agent-written, so end-to-end, property and differential tests are what let it run many agents at once.
  • Linear rebuilt code review for agent-written changes, and Grapim traces each PR line back to the agent run that wrote it.
  • Open-source ARTEX attack agents hit Korean banks and a ConnectWise supply chain; exposed instances grew from 3 to 1,255 since June.
  • TrustGraph and SASY gate agent actions on everything the agent read, not just what it may access.
  • Evals that diff the whole database after a computer-use task catch collateral damage and false completion.
  • Alibaba's Qwen4-Max cuts API prices while DeepSeek rations free-tier compute.

Verification was the day's engineering thread. The Claude Code team and Linear both described how they trust mostly agent-written code: tests that produce evidence, and review rebuilt around agent runs. New eval tools follow the same rule, grading database diffs and repeated trials instead of an agent's own report.

Security made it urgent. ARTEX, an open-source attack-agent framework, is now behind real bank and supply-chain breaches, and policy engines like TrustGraph and SASY gate actions on what an agent read.

Prefer to watch?

LLM Digest Daily Explainer: Nov 16, 2026

How today's stories work, in 1:57. Loads from YouTube when you press play.

Coding agents: verification becomes the bottleneck 4 items

Teams shipping mostly agent-written code are moving trust from line-by-line human review to verification loops and run-level provenance.

How the Claude Code team thinks about verification

Claude Dev BlogDetails

Claude Code team (~90% of its code written and reviewed by Claude) explains how end-to-end, property and differential tests let agents prove their work, and how it keeps those verifiers honest.

Agent evals: grade what the agent did, not what it said 4 items

New eval tooling holds the model fixed, repeats trials and inspects real side effects instead of trusting an agent's own completion report.

Agent permissions: policy that follows the data 5 items

Read-permission checks are not enough: several projects now gate an agent's actions on everything it read before acting.

Releasing Personal Agent Protocol 1.0

Hacker News AIDetails

Sierra ships v1.0 of Personal Agent Protocol (Poppy), a standard for personal agents to sign users in and act through company agents, sites and APIs, with partners including Shopify, Stripe and Walmart.

Security: offensive agents are in the wild 4 items

Autonomous attack tooling has moved from GitHub demos to real campaigns, and open-weight models are closing the hacking gap.

Inference and compute: cheaper paths to serving agents 5 items

Serving work focuses on running many agents on less hardware and on measuring real cost per task across providers.

Model market: price cuts, rationing and token data 3 items

Chinese labs are competing on price while compute limits bite, and Databricks publishes usage data to guide model choice.

You are caught up for this edition